Titulo Estágio
LLM-Augmented Evolutionary Computation
Áreas de especialidade
Sistemas Inteligentes
Local do Estágio
CMS-CISUC
Enquadramento
The rapid development of Large Language Models (LLMs), such as GPT and LLaMA, has opened new possibilities for hybrid AI systems that combine connectionist and evolutionary approaches. In this context, the integration of LLMs into Evolutionary Computation (EC) frameworks offers promising new directions for enhancing evolutionary search, representation, and evaluation processes.
This internship focuses on LLM-augmented Evolutionary Computation, where LLMs are used not as primary solvers, but as embedded tools that support, guide, or enrich evolutionary algorithms. Possible roles for LLMs include generating initial populations, designing variation operators, providing priors for fitness shaping, acting as surrogate evaluators, or introducing domain-specific knowledge in a flexible and scalable manner.
The goal is to explore and prototype different integration strategies, evaluate their effectiveness in selected machine learning or neuroevolution tasks, and contribute to the growing field of Evolutionary Machine Learning powered by foundation models.
Objetivo
• Analyze current research in LLM-augmented EC and related hybrid approaches
• Design and implement mechanisms where LLMs assist or steer evolutionary algorithms, such as:
-- Prompt-based population initialization
-- Language-informed mutation/crossover
-- Language-informed genotype to phenotype mapping
-- Fitness approximation or multi-objective guidance using LLM embeddings or evaluations
• Apply the developed methods to benchmark problems, such as symbolic regression, neural architecture search, or behaviour synthesis
• Evaluate the impact of LLM integration on performance, diversity, and interpretability
• Write a scientific papers reporting results and insights
Plano de Trabalhos - Semestre 1
• Literature review on Evolutionary ML and LLM integration strategies
• Familiarisation with frameworks for EC and LLMs
• Design of LLM–EC hybrid strategies
• Prototyping and controlled experimentation
Plano de Trabalhos - Semestre 2
• Full implementation of selected LLM–EC strategies
• Benchmarking and comparative evaluation
• Analysis of impact on convergence, diversity, and robustness
• Integration of findings into a dissertation and scientific publications
Condições
• The internship will take place at the CDV Lab (CMS-CISUC).
•A research grant for graduates may be provided for a minimum of 3 months, renewable upon agreement. The grant will follow the FCT’s guidelines for graduate fellowships.
Observações
Orientadores:
• Penousal Machado
• Pedro Abreu
Orientador
Cognitive and Media Systems Group - CISUC
penousal@gmail.com 📩